单通道ICA及其在变形分析中的应用
发布时间：2019-01-04 12:16
【摘要】：从变形监测数据中分离出噪声、系统误差、不同影响因素引起的变形或不同特征的变形信息是变形分析的一项主要工作,可为变形预报或工程结构损伤状态识别提供必要的信息。独立分量分析(Independent Component Analysis,简称ICA)是一种盲信号分离方法,在信号分离方面具有独特的优势,可利用原信号的相互独立特性从混合观测信号中分离出原信号,该特性使得ICA可作为变形分析的有力工具。然而独立分量分析要求输入信号不小于输出信号,而在单测点变形分析中通常只有一个通道的观测数据。为此,本文将单通道ICA方法引入变形分析中,在对已有的单通道ICA算法进行比较分析的基础上,深入研究了基于相空间重构的单通道ICA算法,提出了基于增强经验模式分解的单通道时变ICA算法,并利用这些算法进行变形监测数据去噪、变形效应分量的分解及变形建模。主要研究工作与成果如下： 1.研究了独立分量分析的基本理论研究,总结分析了FastICA算法、多维独立分量分析等几种常用的独立分量分析算法。 2.通过无噪和有噪模拟信号实验比较分析了动态嵌入ICA(DE-ICA)、小波ICA (WT-ICA)算法、经验模式分解ICA (EMD-ICA)和增强经验模式分解ICA(EEMD-ICA)四种单通道ICA算法,结果表明：DE-ICA算法的信号分离效果最优；EMD-ICA算法最为简单,无需任何参数设置；WT-ICA算法和DE-ICA算法速度较快,EMD-ICA算法其次,而EEMD-ICA算法由于多次迭代运行效率最低。 3.利用相空间重构方法选取合适的嵌入维数和时间延迟,解决DE-ICA中嵌入维数过多会导致信号冗余和难以选取的问题,形成了基于相空间重构的单通道ICA算法(PSR-ICA)。模拟实验验证了相空间重构的单通道ICA算法不仅可以数据去噪,而且可以有效地分离观测信号中的独立信号。 4.利用PSR-ICA对五强溪大坝位移监测数据进行了分析,首先利用PSR-ICA对位移监测数据去噪,然后对利用该方法分离的位移分量与温度、水位等影响因素进行关联分析。结果表明：其中两个提取的位移分量分别与温度和水位高度相关,与利用大坝统计位移模型计算的温度和水位分量也基本一致,说明了PSR-ICA可有效地分离出大坝位移的各效应分量。 5.研究了在线盲源分离的ICA算法——EASI算法,该算法可用于识别混合矩阵缓慢变化,具备FastICA这类离线算法所没有的时变处理能力。模拟实验验证了EASI算法在多通道的时不变、突变和时变混合情况下的良好分离效果。提出了基于EEMD的单通道EASI算法(EEMD-EASI),并将其应用于五强溪大坝位移监测数据的单通道时变ICA分离中,结果表明:EEMD-ICA能够更准确地分离出温度效应位移分量及时效分量,从时变混合矩阵可以看出大坝蓄水几年后趋于稳定。
[Abstract]:The main work of deformation analysis is to separate out the deformation information caused by noise, systematic error, different influencing factors or different characteristics from the deformation monitoring data. It can provide necessary information for deformation prediction or damage state identification of engineering structures. Independent component Analysis (Independent Component Analysis,) is a blind signal separation method, which has a unique advantage in signal separation. The original signal can be separated from mixed observation signal by using the independent characteristics of the original signal. This feature makes ICA a powerful tool for deformation analysis. However, independent component analysis (ICA) requires that the input signal is not less than the output signal, but in the deformation analysis of a single measurement point, there is usually only one channel observation data. In this paper, the single-channel ICA method is introduced into the deformation analysis. Based on the comparison and analysis of the existing single-channel ICA algorithm, the single-channel ICA algorithm based on phase space reconstruction is deeply studied. A single channel time-varying ICA algorithm based on enhanced empirical mode decomposition is proposed, and these algorithms are used to Denoise the deformation monitoring data, decompose the deformation effect components and model the deformation. The main research work and results are as follows: 1. The basic theory of independent component analysis (ICA) is studied, and several common independent component analysis (ICA) algorithms, such as FastICA algorithm and multidimensional ICA algorithm, are summarized and analyzed. 2. Four single channel ICA algorithms, including dynamic embedded ICA (DE-ICA), wavelet ICA (WT-ICA), empirical mode decomposition (EMD-ICA) and enhanced empirical mode decomposition (ICA (EEMD-ICA), are compared with each other in noise-free and noise-free analog signal experiments. The results show that the DE-ICA algorithm has the best signal separation effect. The EMD-ICA algorithm is the simplest without any parameter setting. The WT-ICA algorithm and the DE-ICA algorithm are faster than the EMD-ICA algorithm, while the EEMD-ICA algorithm is the least efficient because of multiple iterations. 3. The phase space reconstruction method is used to select the appropriate embedding dimension and time delay to solve the problem that too many embedded dimensions in DE-ICA will lead to signal redundancy and difficult to select. A single channel ICA algorithm (PSR-ICA) based on phase space reconstruction is formed. The simulation results show that the single-channel ICA algorithm for phase space reconstruction can not only de-noise the data, but also effectively separate the independent signals from the observed signals. 4. The displacement monitoring data of Wuqiangxi dam are analyzed by PSR-ICA. Firstly, the displacement monitoring data are de-noised by PSR-ICA, then the displacements separated by this method are correlated with temperature, water level and other influencing factors. The results show that two of the extracted displacement components are related to temperature and water level respectively, and are consistent with the temperature and water level components calculated by using the dam statistical displacement model. It is shown that PSR-ICA can effectively separate all effect components of dam displacement. 5. In this paper, the ICA algorithm of online blind source separation, EASI algorithm, is studied. This algorithm can be used to identify the slow change of the mixed matrix, and has the ability of time-varying processing without the off-line algorithm such as FastICA. The simulation results show that the EASI algorithm has good separation performance in the case of multi-channel time-invariant, abrupt and time-varying mixing. A single channel EASI algorithm (EEMD-EASI) based on EEMD is proposed and applied to the separation of single channel time-varying ICA from the displacement monitoring data of Wuqiangxi Dam. The results show that EEMD-ICA can more accurately separate the temperature effect displacement component and the aging component, and from the time-varying mixing matrix, it can be seen that the dam tends to be stable after several years of water storage.
【学位授予单位】：中南大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TU196.1
本文编号：2400279
[Abstract]:The main work of deformation analysis is to separate out the deformation information caused by noise, systematic error, different influencing factors or different characteristics from the deformation monitoring data. It can provide necessary information for deformation prediction or damage state identification of engineering structures. Independent component Analysis (Independent Component Analysis,) is a blind signal separation method, which has a unique advantage in signal separation. The original signal can be separated from mixed observation signal by using the independent characteristics of the original signal. This feature makes ICA a powerful tool for deformation analysis. However, independent component analysis (ICA) requires that the input signal is not less than the output signal, but in the deformation analysis of a single measurement point, there is usually only one channel observation data. In this paper, the single-channel ICA method is introduced into the deformation analysis. Based on the comparison and analysis of the existing single-channel ICA algorithm, the single-channel ICA algorithm based on phase space reconstruction is deeply studied. A single channel time-varying ICA algorithm based on enhanced empirical mode decomposition is proposed, and these algorithms are used to Denoise the deformation monitoring data, decompose the deformation effect components and model the deformation. The main research work and results are as follows: 1. The basic theory of independent component analysis (ICA) is studied, and several common independent component analysis (ICA) algorithms, such as FastICA algorithm and multidimensional ICA algorithm, are summarized and analyzed. 2. Four single channel ICA algorithms, including dynamic embedded ICA (DE-ICA), wavelet ICA (WT-ICA), empirical mode decomposition (EMD-ICA) and enhanced empirical mode decomposition (ICA (EEMD-ICA), are compared with each other in noise-free and noise-free analog signal experiments. The results show that the DE-ICA algorithm has the best signal separation effect. The EMD-ICA algorithm is the simplest without any parameter setting. The WT-ICA algorithm and the DE-ICA algorithm are faster than the EMD-ICA algorithm, while the EEMD-ICA algorithm is the least efficient because of multiple iterations. 3. The phase space reconstruction method is used to select the appropriate embedding dimension and time delay to solve the problem that too many embedded dimensions in DE-ICA will lead to signal redundancy and difficult to select. A single channel ICA algorithm (PSR-ICA) based on phase space reconstruction is formed. The simulation results show that the single-channel ICA algorithm for phase space reconstruction can not only de-noise the data, but also effectively separate the independent signals from the observed signals. 4. The displacement monitoring data of Wuqiangxi dam are analyzed by PSR-ICA. Firstly, the displacement monitoring data are de-noised by PSR-ICA, then the displacements separated by this method are correlated with temperature, water level and other influencing factors. The results show that two of the extracted displacement components are related to temperature and water level respectively, and are consistent with the temperature and water level components calculated by using the dam statistical displacement model. It is shown that PSR-ICA can effectively separate all effect components of dam displacement. 5. In this paper, the ICA algorithm of online blind source separation, EASI algorithm, is studied. This algorithm can be used to identify the slow change of the mixed matrix, and has the ability of time-varying processing without the off-line algorithm such as FastICA. The simulation results show that the EASI algorithm has good separation performance in the case of multi-channel time-invariant, abrupt and time-varying mixing. A single channel EASI algorithm (EEMD-EASI) based on EEMD is proposed and applied to the separation of single channel time-varying ICA from the displacement monitoring data of Wuqiangxi Dam. The results show that EEMD-ICA can more accurately separate the temperature effect displacement component and the aging component, and from the time-varying mixing matrix, it can be seen that the dam tends to be stable after several years of water storage.
【学位授予单位】：中南大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TU196.1
【参考文献】
相关期刊论文 前2条
1 徐洪钟,吴中如,李雪红,施斌;基于小波分析的大坝变形观测数据的趋势分量提取[J];武汉大学学报(工学版);2003年06期
2 伍法权，，王尚庆;卡尔曼滤波方法在链子崖危岩体变形适时预报中的应用[J];中国地质灾害与防治学报;1996年S1期
本文编号：2400279
本文链接：https://www.wllwen.com/guanlilunwen/chengjian/2400279.html
